Back
[09:51:36] <KGB-linuxcnc> 03Chris Radek 052.6 7ac318c 06linuxcnc 10src/emc/rs274ngc/rs274ngc_pre.cc Log interp messages to stderr as intended, instead of crashing * 14http://git.linuxcnc.org/?p=linuxcnc.git;a=commitdiff;h=7ac318c
[09:53:25] <cradek> well I couldn't figure out the weird remap problem, so I fixed this trivial bug instead
[10:54:43] <seb_kuzminsky> heh
[11:43:35] <kwallace> Hello. I'm looking at arbitrary 2D probing and 3D probing. I would like to find out what the current status is to see what needs to get done to make it more mainstream.
[11:44:53] <seb_kuzminsky> kwallace: i think it just works, G38,
http://linuxcnc.org/docs/2.6/html/gcode/gcode.html#sec:G38-probe
[11:45:26] <cradek> I think I don't understand what you mean
[11:45:37] <cradek> say more?
[11:51:22] <kwallace> I have routines I made for probing rectangular and circular objects which result in finding their center. I would like to go further by digitizing 2D outlines and 3D surfaces, but I'm hoping to not have to start from scratch.
[11:52:01] <cradek> http://linuxcnc.org/docs/2.5/html/examples/gcode.html#_smart_probe
[11:52:09] <cradek> smartprobe.ngc in the distro is pretty neat
[11:52:36] <cradek> it tries to minimize the motion used to probe a grid
[11:53:09] <cradek> probing around an arbitrary 2d shape sounds hard but cool
[11:53:36] <cradek> once you have that, you can do 3d waterline probing
[12:00:22] <kwallace> Probing a circular boss turned out to be harder than expected. I do three narrow probes in X to get a very rough calculation of the center of an arc. Then if the arc is fairly shallow I do another wider scan to get a more accurate center. The final scan is on X and Y centers.
[12:00:27] <kwallace> I need the narrow scans in case I'm probing a small pin which the wide scan would miss. The narrow scan seems to be too course to get an accurate center. As the shape becomes less known the factors I need to consider increase.
[12:01:46] <kwallace> Hmm, I found this too,
http://wiki.linuxcnc.org/cgi-bin/wiki.pl?ProbeKins
[12:03:48] <cradek> probing around an arbitrary 2d shape is an interesting geektrap
[12:05:07] <cradek> it seems like you can maintain an approximate edge normal perpendicular to the previous two points
[12:05:35] <cradek> and just walk around the shape with it
[12:06:21] <cradek> sharp inside corners will be tricky
[12:06:36] <kwallace> https://www.youtube.com/watch?v=RZGP7zNWq-I
[12:07:44] <cradek> that looks like smartprobe's algorithm
[12:08:00] <cradek> very weird that the probe tip isn't round
[12:11:37] <pcw_home> Video aspect ratio problem?
[12:14:38] <cradek> oh maybe so
[12:16:38] <jepler> the video area seems to have an unusual ratio of ~1.2:1. scaling the frame so it has a 16:9 ratio, the probe tip is more spherical
[12:16:51] <jepler> but the thing being probed no longer looks like a huge fake molar
[12:22:32] <kwallace> Thank you for the link, I have some homework to do.
[12:25:01] <pcw_home> A molar? I though it was a sculptural rorschach
[12:47:46] <kwallace> Corrected view? :
http://wallacecompany.com/tmp/Screenshot_probing_on_steroids-1b.png
[13:33:36] <CaptHindsight> jepler: new version $74 odroid-xu4 board but ethernet is over USB leaving only SPI for real time IO
http://linuxgizmos.com/79-dollar-octacore-sbc-runs-ubuntu-and-android/
[13:34:48] <CaptHindsight> jepler: how is your new ARM board working out for you?
[13:37:34] <jepler> CaptHindsight: the shipped kernel has flaky wifi and no drivers for usb wifi. self-built kernel doesn't boot properly -- can reach init=/bin/sh but not succeed in starting to text-mode with network enabled
[13:38:05] <jepler> so not much is going on with it at the moment; the hostmot2 ethernet stuff has been on my mind much more
[13:38:30] <jepler> linuxcnc does run and pass its testsuite but without the ability to build any kernel I can't try building an rt kernel
[13:38:46] <CaptHindsight> :(
[13:39:27] <jepler> I know I built my kernel with a different compiler than they did, because I decided to self-host the kernel build on the device..
[13:39:43] <jepler> they are both 4.9.x versions but may have different patches applied to the compiler
[13:47:59] <CaptHindsight> memleak has the cubie2 board (A20) running Linuxcnc but we still have to get a Mesa ethernet board to test hm2_eth
[13:48:21] <CaptHindsight> is more out of curiosity
[13:49:11] <CaptHindsight> the imx6 parts seem to be working best with RT linux so far
[13:49:17] <jepler> yeah, anything arm is just curiosity to me. I like to know we are ported to ARM in general but none of the arm hardware I am familiar with is anything I'd promote for use.
[13:49:20] <jepler> afk
[14:02:51] <micges> CaptHindsight: can you point me to his kernel for cubie2? or at least .config?
[14:09:53] <CaptHindsight> micges: I'll have him ping you with it when he's around
[14:10:33] <micges> thanks
[15:00:38] <skunksleep> If the splash screen doesn't load because of missing axis. (2 axis config loading the axis splash gcode). File - open doesn't work anymore. You can load recent files though
[15:01:42] <seb_kuzminsky> weird
[15:01:56] <seb_kuzminsky> there's a special splash screen for 2 axis machines
[15:02:17] <seb_kuzminsky> (but what you describe sounds like an annoying bug in axis)
[15:11:11] <andypugh> Back home now. Did anyone get anywhere with the carousel_demo bug? My feeling is that the bug isn’t in my code, but i would like confirmation?
[15:12:17] <seb_kuzminsky> andypugh: there was some debugging, it seems to be in the remap code in the interpreter, not in your code, but no resolution was found
[15:13:01] <andypugh> Is anyone on good enough terms with mah to prod him to have a look?
[15:15:46] <skunksleep> cradek:
https://www.google.com/url?sa=t&source=web&rct=j&url=http://www.tormach.com/uploads/899/SB0046_PathPilot_Release_Notes_0315A-pdf.html&ved=0CCMQFjABahUKEwit8_KZxfTGAhWMdD4KHc0GAWE&usg=AFQjCNHTB9JGYg6j-I2Ab0I8nxuCyK7Bfw&sig2=kVy9ZQt2f1ntVwi5j6oNxw
[15:16:52] <skunksleep> Fix for #564 Mill - Interpreter: Spurious duplicate O word error message on M6 MDI command.
[15:17:35] <andypugh> Oooh, that looks like the one...
[15:17:55] <skunksleep> Maybe get with Tormach?
[15:19:06] <seb_kuzminsky> my pathpilot source is from 2015 Feb 23, so it predates that release note
[15:19:20] <skunksleep> I could email rogge and see if we could get a diff
[15:21:23] <andypugh> I already did :-)
[15:21:31] <seb_kuzminsky> they will surely send us their linuxcnc changes
[15:21:33] <seb_kuzminsky> thanks guys
[15:21:35] <andypugh> (Actually I asked them to push the fix)
[15:21:42] <seb_kuzminsky> perfect
[15:21:58] <skunksleep> Awesome
[15:22:11] <seb_kuzminsky> i bet the bug is in 2.6 too, andypugh have you tried your carousel branch (or at least the remap part) there?
[15:22:57] <skunksleep> It was Google.. I searched for the error
[15:23:27] <andypugh> I haven’t tried it, but the config is a standalone thing except for the carousel comp, so it’s easy to test.
[15:26:59] <skunksleep> andypugh: it only happens in mdi?
[15:29:04] <andypugh> I haven’t tried it from G-code. That’s a rather obvious thing that I failed to try.
[15:35:55] <seb_kuzminsky> lovely, we have two different files named stdglue.py, one for gmoccapy and one for everything else
[15:39:57] <andypugh> I can’t help feeling that Norbert might need a few hints about robutsmess and maintanabililty.
[15:43:35] <PCW> hm2_eth up for 3 days now with 2 cards at 3 KHz
[15:43:36] <PCW> max servo thread time = 190 usec nominal=148 usec
[15:43:43] <jepler> hm, this may be the root problem behind the networking weirdness I've seen on the dragonboard
https://www.96boards.org/forums/topic/410cubuntu-cant-ssh-in/
[15:46:49] <skunksleep> Did we figure out the apparent halui slowdown?
[15:47:08] <jepler> skunksleep: don't think so
[15:47:25] <cradek> it's probably not halui - lair is going to bisect it when he can
[15:57:04] <skunksleep> Rob still has one issue to look at. Won't get to it until next week
[16:18:36] <JT-Shop> good that gives me a week to run a comb over the docs
[16:32:14] <seb_kuzminsky> i'd rather take it slow and get it right, than rush and fuck it up
[16:32:21] <jepler> I want to pass my new version of the arm memory ordering patches under seb's nose, as well as the guy who was helpful on the thread
[16:32:43] <seb_kuzminsky> i'll gladly look at it
[16:33:38] <jepler> seb_kuzminsky: branch jepler/arm-atomic
[16:33:52] * seb_kuzminsky fetches
[16:36:06] <jepler> I sort of suspect that fifo->out also needs to be given the atomic treatment
[16:36:08] <seb_kuzminsky> have you tested both with & without stdatomic.h available? (old & new platforms)
[16:36:56] <jepler> yes, it built on odroid u3 (no gcc 4.9), on debian jessie (gcc 4.9), and presumably on the buildbot
[16:37:03] <seb_kuzminsky> ossom
[16:37:24] <jepler> one problem with my system of #ifs is that it won't detect what version of clang, if any, supplies <stdatomic.h> without requiring -std=c11
[16:37:47] <jepler> afk
[16:39:59] <seb_kuzminsky> those two commits look fine to me
[16:57:14] <jepler> hm on reflection I bet ->out also needs to be atomic
[16:58:04] <jepler> thread 1: read last bit of data from sample, increment ->out thread 2: check for space via ->out, write first bit of next sample
[16:59:14] <jepler> if the increment can be moved before the read, then the wrong (newly overwritten value) could be read
[17:08:10] <jepler> a much less likely bug to catch by looking, since it would only occur when running with a full queue, as opposed to an empty one
[17:11:01] <jepler> (I mean, even if you go looking for it to occur)
[17:17:54] <jepler> PCW: just for fun I started 3 cards at 2kHz on my core2.
[17:18:25] <jepler> seb_kuzminsky: oh and there were more hm2 eth commits, but I'm not sure the branch is in final shape. after I went looking for the reason for intermittent startup problems, I found bugs..
[17:19:26] <seb_kuzminsky> i thought skunkworks found the last bug the other day
[17:19:33] <jepler> origin/2.7..jepler/hm2-eth-startup
[17:19:36] <seb_kuzminsky> who ordered this one?
[17:19:53] <jepler> oh this is the one we were talking about, I think
[17:20:06] <seb_kuzminsky> sure
[18:00:11] <PCW> jepler: I'll try 4 cards when I get a chance, sort of curious what the additional per card
[18:00:13] <PCW> overhead is, assuming similar packet sizes. It does not seem like much
[18:08:35] <seb_kuzminsky> andypugh: 2.6 has the same remap oword problem, fyi
[18:08:45] <seb_kuzminsky> yay, weekend
[18:08:46] <seb_kuzminsky> bbl
[18:24:11] <andypugh> seb_kuzminsky: I thought it would have.
[18:24:42] <andypugh> (Not really all there tonight, back home after 2 weeks working in Granada)
[18:42:52] <KGB-linuxcnc> 03John Thornton 052.7 2f96681 06linuxcnc 10docs/src/gui/gmoccapy.txt 10docs/src/gui/touchy.txt Docs: add links to basic hal * 14http://git.linuxcnc.org/?p=linuxcnc.git;a=commitdiff;h=2f96681
[21:21:54] <jepler> Period FP Name ( Time, Max-Time )
[21:21:55] <jepler> 500000 YES thread1 ( 482363, 580925 )
[21:22:33] <jepler> running 3 cards for a few hours at 2kHz. Typical CPU usage is 15% rtapi_app, 5% rx and 5% tx interrupts
[21:25:26] <jepler> this is just hal, not linuxcnc so no "task" in the thread
[21:50:36] <cradek> at least one run took longer than the nominal period?
[21:50:45] <cradek> what happens then?
[22:51:41] <pcw_home> period is in ns, time is in CPU clocks
[22:59:30] <cradek> heh I thought I knew what "time" means
[23:02:46] <pcw_home> Ive gotten use to dividing by the CPU clock
[23:15:10] <seb_kuzminsky> haha, an embedded computer called the 'EMC2-Z7015'
[23:15:13] <seb_kuzminsky> http://linuxgizmos.com/stackable-sbc-sandwich-runs-linux-on-armfpga-zynq-soc/
[23:15:24] <seb_kuzminsky> we^Wsomeone should sue them
[23:16:06] <cradek> weird name
[23:16:13] <cradek> fwiw, I like our new name better
[23:16:26] <cradek> (doesn't mean I am glad it happened that way)
[23:16:36] <seb_kuzminsky> "below $800 for 100+ volumes"
[23:16:52] <seb_kuzminsky> the new name is better in most ways
[23:17:07] <seb_kuzminsky> i miss sometimes only typing 4 letters
